Reconstruction of the full spectrum of solar-induced chlorophyll fluorescence: Intercomparison study for a novel method

Abstract

Solar-Induced chlorophyll Fluorescence (SIF) can serve as an early and non-invasive indicator of the functioning and status of vegetation due to its close link to photosynthetic activity. Most existing approaches retrieve SIF at around few discrete absorption lines. However, the full SIF spectrum can provide more information on the functional status of photosynthetic machinery. European Space Agency's FLuorescence EXplorer (FLEX) mission, to be launched in 2022, is dedicated to the accurate reconstruction of the full SIF spectrum over land and incorporates the heights and positions of the two SIF peaks and the total fluorescence emission (spectrally-integrated value) into planned Level-2 products.In this paper, an advanced Fluorescence Spectrum Reconstruction (aFSR) method was proposed to reconstruct the full SIF spectrum by capitalizing on the features of existing methods. The aFSR method used linear combinations of basis spectra to approximate the spectra of SIF and the reflectance factor and exploited all available bands within the spectral range of SIF emission for spectral fitting of SIF and reflected radiance. The number of basis spectra of the reflectance factor used was self-adaptively determined based on the Bayesian information criterion. A comprehensive intercomparison between the aFSR method and three other methods (i.e., the Fluorescence Spectrum Reconstruction method, the Full-spectrum Spectral Fitting Method, and the SpecFit method) was performed using simulated and experimental datasets. For simulated datasets, the impact of spectral resolution (SR), signal-to-noise ratio (SNR), atmospheric correction, canopy structure, leaf biochemical parameters and directional effect on the accuracy of SIF spectrum reconstruction was considered. Results show that while all methods could achieve the accuracy standard set by the FLEX mission (average absolute relative error of spectrally-integrated SIF <10%) when spectral resolving power and SNR were high (e.g., SR ≤ 0.3 nm and SNR ≥ 700), aFSR generally provided the highest reconstruction accuracy. For the first time we investigated the performance of the SIF spectrum reconstruction on 3-D radiative transfer (RT) simulations and compared with that on typical 1-D simulations. The increase of canopy heterogeneity from 1-D to 3-D did not noticeably deteriorate the accuracy of aFSR, implying that aFSR was applicable to different canopy structures. The aFSR method was also more robust than other methods as it was less affected by atmospheric correction and directional effect. For the experimental dataset, the SIF spectra reconstructed by aFSR agreed well with literature in terms of shape, magnitude and diurnal variation and were in agreement with the other methods: the coefficient of determination and the root-mean-square error between the reconstruction results of aFSR and the average of the SIF spectra reconstructed through three other methods were higher than 0.93 and lower than 0.09 W·m−2·sr−1·μm−1, respectively.